Re: Speed of user-defined aggregates using array_append as transfn

I wrote:
> I'm still worried about chewing up a kilobyte-at-least for each transition
> value, but maybe that's something we could leave to fix later. Another
> idea is that we could teach the planner to know about that in its hash
> table size estimates.

Here's a complete proposed patch for this. I decided to hard-wire the
planner adjustment to apply to array_append specifically. One could
consider doing it whenever the aggregate transtype is an array type,
but that seems likely to be casting too wide a net for now. We can
revisit it in the future if necessary. In any case, the estimate
can be overridden per-aggregate using the aggtransspace parameter.

Barring objections, I intend to back-patch this as far as 9.5.

regards, tom lane